Some will say “finally!” After years of waiting for Rockchip RK3588 processor, ROCKPi Trading Limited/Radxa got some samples for their ROCK5 Model B single board computer and has started to take pre-orders with discounted prices starting at $79 through distributors.
But let’s check out the specifications first, with the octa-core Cortex-A76/A55 Pico-ITX SBC shipping with up to 16GB RAM, M.2 NVMe storage, 2.5GbE, optional WiFi 6E, 8K video output via HDMI or USB-C ports, 4K HDMI input, and more.
Radxa ROCK5 Model B (aka ROCK 5B) specifications:
- SoC – Rockchip RK3588 octa-core processor with four Cortex-A76 cores @ 2.4 GHz, four Cortex-A55 cores @ 1.8 GHz, an Arm Mali G610MC4 GPU, a 6TOPS NPU, 8K 10-bit decoder, 8K encoder
- System Memory – 4GB, 8GB, or 16GB LPDDR4x
- Storage
- M.2 2280 socket for NVMe SSD (PCIe 3.0 x4) up to 2,000 MB/s
- MicroSD card socket
- eMMC flash socket
- Video Output
- 2x HDMI 2.1 up to 8Kp60
- 1x USB-C via DisplayPort alt. mode up to 8Kp30
- Three independent displays supported
- Video Input
- 1x micro HDMI input up to 4Kp60
- 2x MIPI CSI connectors
- Networking
- 2.5 Gbps Ethernet RJ45 port with PoE support
- Support for WiFi 6E and Bluetooth 5.2 M.2 module
- USB – 2x USB 3.0 Type-A ports, 1x USB 3.0 Type-C port, 2x USB 2.0 ports
- Expansion – 40-pin GPIO header
- Dimensions – 100 x 72mm (Pico-ITX form factor)
The company will provide Android 12 and Debian “Buster” images based on Linux 5.10 LTS for the board. But since Radxa only managed to get hold of RK3588 samples recently, you may imagine it will take time before the software is properly ported to the board and that’s one of the reasons why they have a Q2 2022 shipping target date. Based on th early RK3588 benchmarks we’ve found, Rockchip RK3588 should be over twice as fast as Amlogic S922X as found in ODROID-N2+ for many tasks, and the GPU performance increase should even be more impressive.
So how much does the board cost exactly? Those are the standard prices:
- $129 with 4GB RAM
- $149 with 8GB RAM
- $189 with 16GB RAM
This is getting quite close to Intel hardware, but ROCK5 Model B has some features not found in most platforms at that price including HDMI input, MIPI CSI camera interfaces, GPIO header, and 2.5GbE.
But as mentioned in the introduction you can get the board for as low as $79 by pre-ordering the board. To get this price, you’ll need to pay a $5 deposit (called R3 code “Radxa ROCK5 Redeem”) to reserve the board, and then you’ll be able to get a $50 discount on the prices above, meaning $79 for the 4GB version, $99 with 8 GB RAM, and $139 for the model with 16GB RAM. This is only valid for one board, and the R3 code is refundable at any time before shipping if you decide you don’t want to board anymore.
You should eventually be able to get the board from any Radxa distributors, but at this time only two resellers participate in the program namely Ameridroid and Allnetchina.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
Nice board with nice layout. Hope they include case like N2+. Good price when discounted. As for standard prices, not so much, as Jean said: too close to intel’s price.
Wondering about the Ethernet chip (RTL8125B?) and whether Tom took care that Hardkernel’s ‘H2 Net Card’ would directly fit into the M.2 slot on the bottom.
I would assume the slot can be configured to switch from Gen3 x4 to x1+x1+x1+x1 in u-boot?
Currently we use PCIe 3.0 x4 for NVMe SSD, one PCIe 2.0 for RTL8125BG, one PCIe 2.0 for WiFi6. Yes, PCIe 3.0 x4 can be configured as four x1.
what with third PCIe 2.0 ? it’s sacrifised for USB3.0?
That’s an extremely good price.
Had expected it to be $200+. The first cheap arm SBC that can actually function as a day-to-day PC withoout many compromises.
Actually stronger in CPU horsepower than an N5095:
https://www.cnx-software.com/2021/12/16/rockchip-rk3588-benchmarks-show-up-on-geekbench/
https://browser.geekbench.com/v5/cpu/6665182
> Actually stronger in CPU horsepower than an N5095
Based on what? Doing excercises in stupidity staring at combined Geekbench scores like these https://browser.geekbench.com/v5/cpu/compare/11836399?baseline=11144544
Even when taking another Geekbench number for some unknown RK3588 device like https://browser.geekbench.com/v5/cpu/compare/11371316?baseline=11144544 the only conclusion I would draw from this is that with the tested RK3588 device something’s seriously wrong since the existence of four A55 cores clocking at 1.8 GHz should result in much higher multicore numbers.
The first one has its A76s clocked at 2 ghz.
https://browser.geekbench.com/v5/cpu/11836399
The second link @ 2.2 ghz
https://browser.geekbench.com/v5/cpu/11371316
Final board should be 2.4 ghz.
> I would draw from this is that with the tested RK3588 device something’s seriously wrong since the existence of four A55 cores clocking at 1.8 GHz should result in much higher multicore numbers.
Throttling? Memory bottleneck?
> The first one has its A76s clocked at 2 ghz … The second link @ 2.2 ghz
So if the 1st achieves a 500 ‘single core score’ then at 10% higher clockspeed we must see a 10% better score? Nope, the 2nd link shows 643 which is more than 20% better. Same with multicore scores: 1803 vs. 2397 where the difference is even higher. The scores differ (also) for other reasons.
The only sure thing is: staring at combined Geekbench scores is stupid. 🙂
That’s fair.
On a sidenote. We can stare at GB5 results of comparable chips like the MT6873 or MT6875. Which have plenty of sample size.
But you’re right. Historically arm chips have also always underperformed their geekbench score for me. And I already know that won’t change because neon still sucks.
> Final board should be 2.4 ghz.
Why do you think so at this moment?
Wrt (benchmark) performance: for sure we’ll see better numbers over time due to optimized software/settings at least with OS images based on Rockchip’s BSP (their own forks of older u-boot and kernel releases).
Once communities start to mainline stuff performance may degrade then again…
Well, that post of mine got approved quite late…
Anyway:
> Why do you think so at this moment?
I’ll trust the specsheet is at least reliable… It doesn’t affect me that much. I’m not someone to preorder so if it’s wrong it’s wrong.
Those parts are clocked lower if you look at their individual geekbench page.
> RK3588 device something’s seriously wrong since the existence of four A55 cores clocking at 1.8 GHz should result in much higher multicore numbers.
Throttling? Memory bottleneck?
> Those parts are clocked lower if you look at their individual geekbench page.
Geekbench still trusts in some numbers here and there but does not measure clockspeeds. This crappy TV box based on Amlogic S905 would’ve been reported by Geekbench at running at 2.0 GHz while in reality this thing is stuck at 1.5 GHz.
> Throttling? Memory bottleneck?
That’s the questions a benchmark should answer. Geekbench is the opposite of this. Unfortunately that’s true for almost the whole ‘benchmarking industry’.
> Had expected it to be $200+
Let’s compare the ‘list prices’ or what Allnet China shows with reality (that’s for me in Germany http://www.innet24.de as Allnet outlet for consumers). Radxa Zero D1H: $15,80 vs. 35,19€, EcoPi S housing: $4,99 vs. 8,32€, RockPi 4 RS114B-D2E16: $59,00 vs. 83,54€
The formula seems to be ‘phantasy price in $’ * ‘1.4 up to 2.0’ = real price in €.
‘$129 with 4GB RAM’ then translates to ~180€ which is today ~$205 real price (shipping still not included).
I know it tends to come up higher. Still expected the listed price to be higher than it is.
This will change. We have the plan to manufacture in Europe, it will happen some time in 2023.
This is an excellent initiative! I suspect lots of buyers are indeed there.
I still don’t understand why that is a doubt in 2022. Europe has double the population of US and Canada together but companies artificially keep doing things outside Europe, just because.
It’s same situation of UK, where companies are stubborn to place their a**es in but meanwhile most of their customers are from the “rest of Europe”. Why not to move to any other country other than UK? because it would make sense, and XXI century is the one to make things nonsensical.
> I still don’t understand why that is a doubt in 2022 […] but companies artificially keep doing things outside Europe, just because The problem we all know there is that consumers have been so much told to buy clean and slick finished products that I really suspect that the DIY market is not huge in Europe. And an SBC definitely fits into the DIY branch. We all know many people who don’t even want to refresh their PC with a new CPU/RAM/GPU etc, they prefer to buy a new one because what they have is old and they have… Read more »
Taxes, costs, regulations are higher in Europe, so it may only make sense once a company becomes larger. I remember talking with a French woman in China telling me we would never open a business in France. That was in 2005, and it must be worse now, when I see more regulations on individuals, like regulations/requirements for houses.
I don’t think it’s got worse since 2005. For sure french employees can be annoying to work with or even troublemakers. But in parallel, it’s easy to request financial aids for lots of things related to innovation, or for setting up a business in areas where jobs are rare, or for working on activities around energy savings, so there’s still a balance to find. For example some car manufacturers are proud to say they’re installed here. And I’m pretty sure we’ll see more silicon companies establish here in the upcoming years. Time will tell 🙂
Maybe this? https://www.politico.eu/article/lithuania-taiwan-china-microchip-windfall-clash/
Wow! And that is some news! Please make sure this is heavily announced at least here so security sensitive or china allergic user will not missed that.
The problem is that $ prices are usually without VAT while EU prices are by law (for end consumer) will all VAT included.
Just…no.
Provided that you can’t actually compare 1:1 an ARM to a x86 CPU, you’re talking about Jasper Lake here. Those CPU’s have a completely new fab and are actually very competent for multimedia and office tasks. On the other hand, A76 cores are quite old and I doubt any Linux will perform not even close in therms of snappiness and response in those than in any Jasper Lake CPU.
This board will be very good for many things, but forget about replacing your x86 mini PC with this, sorry.
> standard prices … getting quite close to Intel hardware
And most probably we’ll say the same about consumption figures once people start reviewing the final product in half a year…
Probably.. .Should still be lower, but the N5095 also has more IO…
I think I should buy a router with 2.5GbE and WiFi 6 (and maybe WiFI 6E) this year. Any recommendations? Something within a $200 budget supported by an open-source Linux distro (Debian/OpenWrt) would be nice.
Get a 2.5Gbps switch instead, very few routers with 2.5Gbps, and most only have one port, which is useless.
$120 or so for a switch.
Also, 6E isn’t legal in most countries.
The reason I need it is for testing. If I had a router I’d be able to test iperf, transfer files from fast storage over the 2.5 GbE connection. If I get a switch I’d need an extra 2.5GbE platform beside the DUT, unless I’m missing something.
I’m trying to get that one for review: https://www.cnx-software.com/2021/12/23/dual-10gbe-router-supports-wifi-6-ax3600/, but last time they did not even answer my request for more information, so I’m not hopeful. I won’t be officially supported by OpenWrt either…
Or maybe try to build your own based on a Clearfog or equivalent system, by selecting the components you need/want.
Well, that’s obviously one of few routers with two 10Gbps ports.
Asus has some coming later this year, but I have no idea when.
Routers aren’t alway ideal iperf servers though.
I use my NAS that has an old Core i7 in it and a 10Gbps card.
> it is for testing … If I get a switch I’d need an extra 2.5GbE platform
Well, for testing you need ‘an extra 2.5GbE platform’ anyway since some sort of baseline is needed. What if your single 2.5GbE device is limited by itself and not able to exceed e.g. 1.7Gbps (instead of the 2.32 that are to be expected)?
BTW: RTL8516B USB3 dongles do exist 🙂
That would be a solution. 2.5 GbE switch plus RTL8516B USB 3.0 dongle for laptop. That would work for 2.5 GbE provided the USB3 dongle can achieve 2.32GB… The first DUT would be a Tiger Lake mini PC (UP Xtreme i11 Edge Compute Enabling Kit), so I’d expect good performance from it… But I’d also need extra hardware for WiFi 6, with another dongle that works in access point mode. Not sure how well that would work. I’d prefer an actual WiFi 6 router plus a 2.5 GbE connection. Unless I can safely assume I’d never achieve over 1Gbps real… Read more »
What about mikrotik rb5009? https://forum.openwrt.org/t/add-support-for-mikrotik-rb5009ug/104391
To be honest, I’ve been reading the site hoping that something like this would pop up. For the Wifi, I went with a Belkin RT3200 due to the price and OpenWRT “support”. Wanted to pair it with a 2.5GbE switch, but have been unable to find something that I could cost-justify.
QNAP, TP-Link, Trendnet all have “affordable” 5 and 8-port 2.5Gbps switches.
The list of OpenWRT supported machines that come with WiFi 6 is rather short, and I think none of them come with 2.5GBit ethernet: https://openwrt.org/toh/views/toh_available_16128_ax-wifi
Out of those, I would recommend the MT7622 (Cortex-A53) based ones over the MT7621 (MIPS 1004K) models, unless you need 2.4GHz Wifi-6 in addition to the 5GHz link.
The MT7622 is only dual core though, so a very weak CPU, not that the MT7621 has more compute power, but neither is ideal as a test target.
Wow, a board with decent layout, finally.
Pricing isn’t so great though, but then again, nothing has good pricing at the moment.
The prices in the article are different on the ameriDroid site. The 4GB is listed as 159.99 USD. Are they adding an amount on top of the “standard” price?
And guess what’s written at the other distributor’s site? ‘Because ALLNET China is not able to pre-collect the related VAT on the checkout yet, you will be contacted directly when the parcel arrives in your country and asked to pay the outstanding amount.’
I’m always stunned by people trusting into those ‘list prices’ not containing shipping/taxes/VAT/whatever people really have to pay to get the product into their hands…
Chill please.
> you can get the board for as low as $79 by pre-ordering the board.
LOL! You can pre-order exactly one such board for $79. You can not ‘get’ it for this price though since what you will be charged with in a few months or half a year will include shipping/taxes/whatever.
ROCK5 real prices will be a lot higher since what we as consumers need to pay for holding the product in our hands is not the number displayed in some funny asian web shop…
Okay, we all wish we could actually afford these products, but in all fairness, the mark ups are mostly not the fault of the people making them. Shipping would obviously be less if their factory happened to be near you (or I for that matter) but for better or worse the regulations and taxes in most countries in the developed world could make the total cost higher. As for taxes like on the finished product like sales tax and VAT, well, just like all of the regulations and taxes on the manufacturers, these are our faults for voting for people… Read more »
> the mark ups are mostly not the fault of the people making them Who cares? IMO it is rather weird to talk about irrelevant prices (something listed on some online shop somewhere) while ignoring real prices (what you really have to pay to acquire the product). See the innet24 examples above. This does not only apply to Radxa but to all ‘foreign’ online shops. For example an ODROID M1 with 8GB RAM is listed as a $90 part in Hardkernel’s own shop while here in my location this translates to 120€ (even shipping included for a short time). The… Read more »
I’m from country which have <= 200€ without VAT (not EU) so for me this list price is real. Yes, we have to add delivery but last time it was ~9$ and it is common for internet shops. Sometimes even local ones.
lucky you.. but lets wait for when the product launch for the real price and shipping price.
What’s the recommended power supply?
See the 4-pin header next to the GPIO header. That’s for the $25 ‘ROCK PI 802.3at PoE HAT’ and allows for ~23W which might be a bit on the low side with 2 USB3 consumers.
Then there’s USB-C and USB PD (in the past Radxa also supported Quick Charge 3.0 but IIRC not all steps so this was limited to below 20W)
seen 7W TDP for rk3588, 3x 4.5W for USB3.x, 2x 2.5W USB2, m.2 key-E for WIFI6(E) ~866Mbps ~0.5-2.xW, rtl8125 <0.7W, m.2 key-M SSD(PCIe) <7W (idle ~0.5W, active ~3-3.5W), 2x HDMI <0.5W(?), sdcard <0.35W [SDexpress(PCIe) active <0.7-1.8W], emmc active <0.3W, memory (4-16GB) <3W ~64Gb?
It’s more about knowing, what peripherals highest power needs are. 7W for RK3588 on 5V are (very likely) within 4.6A from a ‘ROCK PI 802.3at PoE HAT’
power supply for PoE: ~44-56V/0.6A, more likely recommended for <=10W (efficiency because of cable length losses vs. local quality power supply)
thx
Thanks but no thanks. I’d rather pre-pay the board at once than pay for a redeem code that I’ll have to remember where I put it in a few months and on which site it’s usable (if at all), to later get a redeem. Also the limit to one coupon per user doesn’t make much sense to me because I could have been interested in pre-ordering a pair. Let’s wait for final board availability then :-/
What would be needed to get an M.2 to mPCIe with 2x Intel gbit i350 adapters?
It works fine on X86_64 with such adapter detected correctly using igb driver.
I am thinking about using that wifi M.2 slot to add 2 additional Gbit ports.
Also very interested in the power draw of this (idle/avg/max)
Idle hopefully lower than an ancient Intel i350-T2 adapter (they waste quite some energy). BTW: RK3588 has two internal GMACs that seem to be routed to nowhere on ROCK5.
Most probably for networking purposes this board is overkill anyway and CM3+ combined with Radxa’s E25 carrier board is a better and less expensive choice…
Ok, if <5W i am fine. I don’t have anything close to it at home.
I need one not only for networking purposes, but multi use cases
Some containers and VMs, NAS (storage via USB3)+++.
I am after something like https://www.lr-link.com/products/lres2202pt.html attached to that WiFi M.2 slot to get 2x Gbit ports
One more thing: how would this SoC compare to an old AMD SoC from t620 plus which is a GX-420CA. Geekbench comparison would indicate ~double the single and treble multicore performace: https://browser.geekbench.com/v5/cpu/compare/11836399?baseline=11987015
The HP t620 plus can use down to ~7w (so i was wrong when i wrote i don’t have anything close to 5W idle power draw) at idle with 256GB mSATA and 2x SODIMM. So it’s quite good low power SoC for my use cases.
> GX-420CA … ~double the single and treble multicore performace
Just take different Geekbench numbers and it looks… differently!
That’s an almost a decade old ’embedded’ AMD G SoC. Of course it will be outperformed by RK3588 which is one of the few multi-purpose ARM designs that balance CPU, GPU and I/O performance/capabilities in a reasonable way. If it wouldn’t be ARM so you can’t run off-the-shelf OS variants but ‘need’ something special.
But there’s some hope. Jared, one of my personal heroes, is porting Tianocore EDK II (UEFI) to RK356x right now (and there’s also some hope RK3588 can benefit from all these RK356x efforts).
While UEFI can be called a mess it allows to avoid more messy stuff like DietPi, Armbian or other ‘ARM distros’ where maintainers include security flaws every now and then just for fun…
Without Armbian, Rockpi 4 would only have kernel 4.4 and the same will happen with this. If you forget what we are doing … All Linux distros for consumers, like this hw is made, is full of shit. This is just the way things are. You get what you pay for.
What are you talking about?
> What are you talking about? This as somewhat recent example. Or check issue #11 in the armbian-config repo (‘Insecure temp file handling’) and then count temp file vulnerabilities in this tool that’s now part of every Armbian install. Manufacturer OS images (or DietPi that bases on those) are of course much more worse since you never know what’s inside while Armbian images are at least created in a controlled fashion so checking for security relevant flaws is somehow possible with a reasonable amount of time and efforts. But if the project ‘owner’ neither understands the meaning of ‘stable’ nor… Read more »
Oh that looks really good! More tempted now.
Yeah, that’s one major downside to deal with these images and being unable to use standard OS installation with such SBCs.
unable to use standard OS installation with such SBCs. All standard OS-es relies on standard raw upstream kernel. Which never gets full support is very poorly maintained. Any functional fix needs months to be implemented, while some other things breaks down. Regardless if you have standard (with more bugs) UEFI mechanism or u-boot. where maintainers include security flaws every now and then just for fun… Of course. You are the expert. Just to tease you and because everything works perfectly fine, something has to be done for our amusement. In open source – if something bothers you, you know what… Read more »
Something like the DeLock D62848 adapter…
Nice board and nice price with the discount. Any word on how the expected PINE64 board will compare? In particular: will it expose SATA or have a second NIC?
How is the ROCK5 using the combo PHYs? Comparing the board specs to chipset specs, maybe for the 2.5 Gbps Ethernet, the wifi/bluetooth M.2 module, and USB 3.0? with the chipset’s dual gigabit Ethernet unused?
Apparently there will be a Khadas board, too. github.com/khadas says “Open Source SBC(Single Board Computer): Amlogic S905X, S912, S905X3, S905D3, S922X, A311D, Rockchip RK3399, RK3588 and more.”
Khadas pricing is quite high AFAIK.
When I look at things rom an overall perspective, at the discounted price, the 4Gb model does look reasonable, even factoring in VAT. Arm based SBC’s and boxes have trundled along at a pedestrian pace compared to what we see in many a mobile phone. So what is actually quite a quantum leap in terms of overall spec, it is welcome, even if 2 years late. Intel continues to screw things up at the bottom end of the SOC chain and whilst the Pi is popular, this trounces it. The Odroid N2+, once you add case, eMMC, RTC and power… Read more »
> Rockchip an their refusal to offer real open source support.
Huh? With which other SoC maker are you confusing RK right now? If RK356x/RK3588 appear soon on http://opensource.rock-chips.com/ (which is to be expected) then devs will get active help from Rockchip staff who also contribute upstream.
> an WiFi 6 options
You know that’s just a single PCIe Gen2 lane (multiplexed with SATA) routed to an M.2 socket?
No confusion. I base my opinion on experience with them and their previous generations of products. I hope that they will do more. Time will tell.
Yes I do know.
> Huh? With which other SoC maker are you confusing RK right now? If RK356x/RK3588 appear soon on http://opensource.rock-chips.com/ (which is to be expected) then devs will get active help from Rockchip staff who also contribute upstream. Quite frankly, for having had to RMA my RK3566-based Station-M2 months ago due to the impossibility to reflash an image in a proprietary format that relied on windows-only proprietary tools, I tend to think that rockchip only opensources devices when they’re forced to (e.g. for chromebooks) or once they get outdated. While the Station-M2 definitely is one of the best device I’ve seen… Read more »
> windows-only proprietary tools
I never understood why you wanted to deal with these tools? Back in 2017 having first encounters with RK SoCs it needed a quick check to realize that using those Rockchip tools is for people flashing the internal storage of Android devices. Easy decision: nothing for me and if a certain device would require these tools (e.g. it smells like a TV box) simply skip the device.
> I never understood why you wanted to deal with these tools I do not! I’m imposed to use them by vendor support 🙁 When some boards are distributed without your OS of choice and the only documented procedure on their site involves either the broken linux variant or the unavailable windows version, you first try the linux variant, which instantly bricks your device, you contact the support and the support asks you to find someone around who still has a windows PC to try the shitty tool, which doesn’t work anymore either once the device is bricked. It’s never… Read more »
> Rockchip staff who also contribute upstream
Do you know what they actually do? And why even previous Rockchip generation is not fully supported in upstream? End users like you – expect fully functional software support without any security flaw. Where we can download that software?
Any information about RK3588 linux mainline patchset ? I know that Android is still on years old linux version 🙂 But mainline would be nice
> I know that Android is still on years old linux version
According to Tom the RK3588 BSP relies on 5.10 which is not ‘years old’ but today not even 13 months.
Wrt Linux mainline there is a rule: it’s ready once the hardware is obsolete. And why/how asking for this now if board makers still have no production silicon but deal with early samples? After years of delays people still buy into announcements? Would you be surprised if ‘Q2 2022’ ROCK5 shipping date will turn into Dec 2022? Me not.
So it looks like turingpi.com is planning compute module with RK3588
That is very interesting. I probably didn’t pay attention to turinpi due to v1 with only RPi module support (i usually ignore all the RPi related media coverage). V2 seems more interesting with support for RK3588, although limited by I/O, IMO my use cases would be very limited with only a single Gb/s port. But i hope we could use mPCIe to 2xGbE adapter on one of the mPCIe slots and get some more network I/O. Combine that with mPCIe to SATA adapters to get 2 additional SATA ports as well. But it will cost you in addition to the… Read more »
I’m wondering if the USB-C port can be re-configured to gadget mode like the Raspberry Pi 4 and Zero. This combined with 4K60 HDMI in creates new opportunity to make an IP-KVM supporting UHD. We already have mature Pi-KVM solutions over there but limited to 1080p due to hardware limitations.
That would be pretty neat. Combine with a relatively cheap multi-port KVM switch if you need multiple devices, and hopefully there are a couple of GPIO’s that you can use to automate the switching.
The price of RK3588 is $55
Quite impossible for that price. Sources?
Tom Cubie is a fraud. His company is going out of business which is why they are doing a pre-order. Don’t pre-order and lose your money.
Yes, don’t preorder or you will lose 5$ 😀
I’m even more amazed that you’ve planned to refund your company by taking $5 at a time 🙂
Sounds great, but without an operating system, it’s useless.
No eDP 🙁
Datasheet shows that HDMI and eDP pins are shared – would it be possible to have eDP output with small (PCB?) adapter?
What’s the connector on the bottom side (opossite of M.2 NVMe slot)? Looks like mini USB, but has too many pins.
That’s the micro HDMI input
too bad AmeriDroid is price gouging and charging $20 over MSRP.
Are you kidding? Don’t you get that AmeriDroid is a distributor/retailer and they run a business (importing goods and doing logistics, warranty handling and so on).
Or do you really think the numbers starting at $129 here in this blog post represent any kind of ‘manufacturer’s suggested retail price’? Seriously?
Their prices are usually as good as it gets if purchasing in the US, it’s not gouging as they are a reseller and have to make a profit.
Their margins are generally very narrow.
The redeem code idea is an excellent way for a small firm like radxa to gauge the market and get some revenue for prototypes.
I don’t think the Mali G610 is supported in Panfrost and expecting working DTS & Drivers to be likely long term.
As a client/desktop this thing has loads of potential, supposedly a really complex SoC though and many have dissapointment of previous and long waits for complete working drivers and image but fingers crossed so put a punt on a $5 redeem code for a 8mb and will see how things go.
There are stress test results, how stable does it hold the maximum load on the LA (CPU, I / O, network)?
BTW: Makefile points to 5.10.66 (Sep 2021) but there are more recent upstream commits included by cherrypicking.
I’ve just checked and unfortunately it’s not 5.10.66, but it goes back to a very old fork of 2.6.32 that lived its own life in an android tree which merged many versions since, up to and including 5.10.66. So it’s not even possible to perform a simple rebase on top of 5.10.66 to get the extra patches, it’s a different kernel with 87515 commits since it forked, who knows how many of which come from mainline, android or various other vendors. To give you an idea, the diff from 5.10.66 is 8.6 MILLION lines, or twice the whole arch+fs directories… Read more »
And keeping it up to date isn’t a trivial task for end-users:
$ git merge v5.10.91
(…)
Automatic merge failed; fix conflicts and then commit the result.
$ git diff|wc -l
1161
$ git diff|grep -c ‘<<<<<‘
46
So there are on average 1.8 conflicts per stable release to be remerged, i.e. you should definitely stick to that kernel tree and not try to upgrade it yourself!
OMFG, forward ported from 2.6.32 on… thank you for checking, saves me doing this myself 🙂
Oh my. Well, this will be fun then 😀
> But I really don’t like these android forks that still pretend to have something to do with linux 🙁
Quote of the day: ‘We will have chips delivered to us in the second half of the year, and rest assured no one else will have them before us‘ 🙂
Followed by “It is unlikely that there will be complete mainline Linux support or working GPU drivers within at least a year from the time that the chips start being delivered.”
Well, this is obvious to anyone familiar with how ‘Linux on ARM’ with consumer devices works though judging by the questions asked in Radxa’s forum they managed to raise consumer expectations in the worst possible way.
IMO Pine64 stepping back from RK3588 early adopters will result in a significant software development delay for the SoC.
With the current state of totally outdated ARM hardware available on linux, it’s likely that intel’s recent (and expensive) chips will have some success in set-top-boxes and small servers. When you see that A53 remains omnipresent 10 years after its release, that the 6-year-old A72 has recently been popularized by RPi and that newer models are still not available, there’s definitely something wrong in this market. The most likely cause is that smartphones are dragging so much money with their closed stuff that it’s pointless to make efforts for the annoying 0.1% left :-/
> the 6-year-old A72 has recently been popularized by RPi
And this move was not related to A72 at all. The significant switch was from VideoCore IV to VI and the software efforts seem to be mostly done (e.g. ThreadX PCIe driver to be able to ‘USB boot’).
If Broadcom just released a VC6 SoC with four A76 guest processors and similar IP blocks compared to BCM2711 then RPi by accident could have the most performant SBC.
Absolutely!
> If Broadcom […] by accident […]
that will almost certainly happen soon(ish) after other vendors ship their A76 sbc’s (like this) en masse.
the rpi market isn’t too shabby for broadcom (about a third of s21 sales), at least in volume.
not to mention the huge mindshare among the dev/diy crowd, which translates into free (for them) mainline kernel support and free marketing with every pi sold.
Hopefully. RPi’s the only one ARM dev SBC that I can get without too much price hiking (and best price/perf too) in my country.
Wrt any RPi with more modern guest processors as Cortex-A72… RPi Trading Ltd. revealed that Broadcom doesn’t do SoCs for them but instead they need to rely on whatever is in BroadCom’s portfolio.
At least me wouldn’t be too surprised by a VideoCore EOL announcement instead of A76 cores on a new BroadCom VC6 SoC.
Please, can you include a description of the powersupply and cable type, passive only and active cooling case types for the Rock 5 Model B? For example, Ameridroid lists the PinePower – 120W Desktop Power Supply but I can’t decide which to get for cabling:
One of the two MIPI connectors is MIPI DSI, not CSI, per the radxa wiki page:
https://wiki.radxa.com/Rock5
Reading up on the Rock5’s Rock5 model B announcement forum thread https://forum.radxa.com/t/introduce-rock-5-model-b-arm-desktop-level-sbc/8361, I get the impression that the SoC is not at all RK3588 but instead a slimmed version RK3588S, which has much less IO: 2x one lane PCIe v2.
The question has been posted here, please have a look: https://forum.radxa.com/t/radxa-rock5-rk3588-pcie-lanes-how-many-to-the-m-2-slots/9580
RK3588S would not be suitable for Rock5. See article and comments @ https://www.cnx-software.com/2022/01/12/rockchip-rk3588s-cost-optimized-cortex-a76-a55-processor/
4 months later and no updates regarding the 5$ promotion
Well, what’s happening with this?
Radxa is sending out a few boards to developers this week for a debug party to check for any potential issues before mass production.
What happens next will probably depend on the results.
Did you get your Rock Pi board to work ?
The ‘debug party’ started: https://forum.radxa.com/t/rock-5b-debug-party-invitation/10483/41?u=tkaiser
Received mine as well (thanks Tom and Thomas BTW). I’ve added my first test results below Thomas’ at the link on the Radxa forum above. TL;DR: the board is amazingly powerful. It’s *really* PC-class performance, and the smarter form factor will definitely help it get adopted widely.
I received mine this weekend, but I have not had time to boot it up yet. I’ll quickly try it today.
Can the early pretesting check seL4 microkernel will be able to port to rock5b rk3588 using odroid-c2 for reference?
Rockchip is using a new RKNPU2 toolkit for the NPU in RK3566/RK3568, RK3588/RK3588S, and RV1103/RV1106.
https://github.com/rockchip-linux/rknpu2
The “old” RKNN Toolkit is only for RK3399Pro(D), RK1806, RK1808, RV1109, RV1126
Looks like Radxa is going to bring out a CM5 based on the rk3588(s) and it’s supposed to be compatible with carrier boards that currently work with their CM3, or the RPi CM4.
Hopefully Radxa’s CM5 will benefit from the work done on the Rock 5B and on Pine64’s QuartzPro64.
Oh and of course the Orange Pi 5
And firefly etc RK3588s boards already out there.
There’s not only the CM5 (compatible to RPi CM4 form factor but with an additional row of 100 pins) but also the NX5 compatible to Nvidia’s SO-DIMM ‘standard’.
And both are based on RK3588S which is just a crippled subset of RK3588. And of course these as well as every other RK3588(S) device will benefit from the work done on RK3588(S) so far since it’s about SoCs and not boards.
Indeed. I wonder whether we’ll see an uncrippled, full RK3588-based CM down the line… I think one of the advantages of SOMs that follow the RPi CM4 form factors is that there are quite a few carrier boards that allow you to continue using cases designed for the RPi3B+/4B. Hopefully there will soon be such carrier boards for SOMs in the DDR2-SODIMM form factor as well.
According to one of Radxa’s forum posts from January, the full RK3588 is too big for a RPi CM4 form factor so I wonder whether it would fit on a RPi CM3 form factor…
> would fit on a RPi CM3 form factor…
…where not even a single PCIe lane is available by a carrier board? The RPi CM ‘standards’ are only suitable for their range of I/O limited Broadcom SoCs and nothing else.
> SOMs that follow the RPi CM4 form factors
…can only expose a single PCIe lane to a carrier board since the CM4 ‘standard’ is designed in a way only suited to the crippled I/O capabilities of their Broadcom BCM2711. That’s why Radxa has another 100 pin connector on their CMs but even with this ‘standard’ using RK3588 would be silly since not able to expose all the additional I/O. You would need another ‘standard’ for RK3588.
For sure it doesn’t eke out the best use of the chip – which is why other carrier boards exist – but keeping it in these form factors does mean that those who previously integrated RPi 3B+/4Bs into their project continue to have potential upgrade paths in terms of flexibility in CPU, RAM and eMMC that doesn’t require them to design and manufacture a whole new case. This means significant reduction in e-waste in, e.g., academic environments.
Radxa provided more details, and I’ve had a look, but sadly backward compatibility may be limited.
https://www.cnx-software.com/2022/09/08/radxa-cm5-rockchip-rk3588s-module-raspberry-pi-cm4/
Thank you for writing this article! I had only heard about the Radxa CM5 yesterday but it looks like there have been forum discussions about it since the first quarter of this year. Thanks also for the update comment that Radxa will keep trying to integrate the Ethernet PHY.
Now advertised for sale on AliExpress
.aliexpress.com/item/1005004697706601.html
and they shamelessly copied the specifications from CNX Software.
“It’s been 1 [year] since you looked at me…”
4GB = $129? More like 4GB = $144, and everytime I’ve gone back to ameridroid’s site, it’s always “pre-order”. (I see 8GB is one that isn’t pre-order)
Reminder for anyone still holding their coupon, it expires March 23 for ameridroid.